fine-scale spatial and temporal variation in earthworm...

Pedobiologia 53 (2010) 127-139

Fine-scale spatial and temporal variation in earthworm surface casting

activity in agroforestry fields, western Honduras

N. Pauli a,b.c.*,1, T. Oberthiir b, E. Barrios a, A.J. Conacher d

• Tropical Soil Biology and Fertility Institute, Centro Internacional de Agricultura Tropical, Colombia

b Land Use Project, Centro Internacional de Agricultura Tropical, Colombia

c Zoological Society of London, Regents Park, London NWI 4RH, UK

d School of Earth and Environment, University of Western Australia, Australia

ARTICLE INFO ABSTRACT

Article history:

Received 19 February 2009 Received in revised form 13 August 2009 Accepted 14 August 2009

Keywords:

Earthworm casts Spatial heterogeneity Temporal variation Organic matter distribution Agroforestry Central America

Quantification of the spatial and temporal distribution patterns of soil fauna is a relatively new area of research, and has been proposed as the key to understanding the high diversity typical of soil fauna communities. Field research on the relationships among the spatial distribution patterns of trees, litter and earthworm surface casting was carried out in two agroforestry fields in a rugged area of western Honduras. Grid-based sampling at a scale of 2-20 m was employed to determine whether any spatial relationships existed among these variables at this fine scale. Each field was sampled twice at either 2 or 3 week intervals, to determine the short-term stability of spatial relationships. The spatial distribution of litter showed a strong pattern of aggregation, whereas earthworm cast distribution did not exhibit strong spatial autocorrelation. However, the spatial distribution patterns of each of these variables were well explained by the spatial arrangement of trees in both sites. Fitted model cross-semivariograms explained between 70% and 90% of the total variation in cross-semivariance between tree density and litter cover, and between tree density and earthworm cast weight. The results of the study suggest that farmers may be able to manipulate populations of earthworms indirectly by managing the spatial arrangement of trees within their crop fields. Planning the distribution of trees could allow farmers to create 'patches' of organic resources within fields, while minimising the negative effects of trees on crop growth due to competition for light, water and nutrients. Over the short time scale of the study, unusually heavy rainfall led to substantial changes in spatial distribution patterns of earthworm cast activity and litter cover, which may otherwise not have occurred. This result emphasises the need to take into account short-term temporal change during ecological studies at fine spatial scales.

Introduction

Once considered as 'noise' that hindered understanding of soil

ecology and biology, the spatial and temporal variability that is

characteristic of many soil macrofauna populations is now recognised as meriting in-depth research (Anderson 1988; Beare

et al. 1995; Barrios 2007). Soil macrofauna, and particularly

earthworms, may have distinct spatial distribution patterns at

distances of less than 100 m. Spatial distribution patterns of soil

macrofauna are likely to be related to population processes such

as dispersal, reproduction and competition, and to the distribu

tion patterns of important resources such as suitable habitat and food sources (Ettema and Wardle 2002; Legendre et al. 2002; Rossi 2003a; Schooley and Wiens 2003). At short time-scales of

• Corresponding author. Current address: Zoological Society of London, Regents Park, London NWl 4RH, UK

E-mail address: [email protected] (N. Pauli).1 Formerly at School of Earth and Environment, University of Western

Australia, Australia.

0031-4056/$-see front matter© 2009 Elsevier GmbH. All rights reserved. doi:10.1016/j.pedobi.2009.08.001

© 2009 Elsevier GmbH. All rights reserved.

days to weeks, soil macrofauna are likely to respond to fluctua

tions in soil temperature, soil moisture and organic resource

distribution.

Earthworms are a dominant component of the soil macrofauna community, and their activities can influence a range of soil

physical, chemical and biological properties, including soil structure, water and gas exchange, organic matter decomposition,

nutrient cycling and availability, and plant growth and crop

yield (Lee and Foster 1991; Lavelle et al. 1997; Six et al. 2004).

In tropical environments, several studies have looked at the finescale spatial distribution of earthworm communities in savanna

environments, by sampling individual earthworm species at

points within a regular sampling grid (Rossi and Lavelle 1998;

Decaens and Rossi 2001; Jimenez et al. 2001: Rossi 2003a, 2003b,

2003c). All of these studies found distinct patterns of aggregation for at least one of the species sampled. The drivers of these

patterns often varied between species. For example, in the savannas of Colombia, Decaens and Rossi (2001) found that larger earthworm species were associated with high root biomass and

total carbon, whereas smaller species were associated with more

128 N. Pauli et al./ Pedobiologia 53 (2010) 127-139

compacted soils with a higher bulk density. In contrast, Rossi

and Lavelle ( 1998) found that the spatial distribution of large

earthworm species at a savanna site in Cote d'Ivoire was close to random, whereas small species had highly aggregated distribution

patterns. Several researchers have looked at temporal variability in soil

fauna populations between seasons and within years. Decaens

and Rossi (2001) sampled earthworms in a tropical pasture in

Colombia at six dates over a period of 1 year, and found that the

clumped spatial aggregation patterns that characterised earth

worm distribution were a constant feature, although the locations

of 'patches' and 'gaps' of earthworms changed over time. Similar

results were reported by Rossi (2003a) for the distribution of earthworms in a region of Cote d'Ivoire. Studies that have sampled soil macrofauna at intervals of 1-3 months have often found large

fluctuations in abundance between sampling dates, related to

seasonal change in rainfall or temperature, or to farm manage

ment practices (Bhadauria and Ramakrishnan 1989; Netuzhilin et

al. 1999; Rossi and Blanchart 2005; Sileshi and Mafongoya 2006).

As many soil macrofauna taxa are highly mobile in comparison

with smaller soil biota, it seems likely that there could be large

short-term fluctuations in their abundance and activity levels.

To date, there has been little research on the variation in spatial

distribution of soil macrofauna over short time periods, such as

days or weeks.

Soil macrofauna distribution may also be influenced by spatial

variability of organic resources, particularly trees. Individual trees

have above- and belowground 'zones of influence' that act as sites

of accumulation for nutrients within the landscape through litterfall, 'scavenging' of nutrients from the surrounding soil by

extensive root networks, and fixation of atmospheric nitrogen by leguminous species (Zinke 1962; Palm 1995; Rhoades 1997;

Casper et al. 2003 ). Trees also affect soil temperature and

moisture regimes through shading, rain interception by canopies,

stemflow and canopy throughfall, and may also influence soil

aggregation and soil texture (Campbell et al. 1994; Rhoades 1997).

In the tropical dry forest region of Mexico, D0ckersmith et al.

( 1999) found that concentrations of plant-available nitrogen and phosphorus were higher in soils near plant stems than in soils

beyond the edge of the canopy dripline. These patterns persisted in the soil for several years after the trees had been removed by

fire. Individual tree species have distinctive root architecture, and

the rhizospheres of different tree species may vary in soil

microbial biomass, enzyme activity and mineralisation rates of

carbon and nitrogen (Phillips and Fahey 2006; Pandey and Palni

2007).

There is evidence to suggest that trees may create localised

areas of high soil biological activity, particularly in savanna

environments where trees are sparsely distributed in the land

scape. Soils beneath trees in African savannas have been

associated with higher soil microbial biomass and greater

nematode concentrations than soil in open areas (Coleman et al.

1991 ), and with greater activity of earthworms and termites,

leading to increased soil macroporosity (Mordelet et al. 1993). In an alley cropping system in Nigeria, earthworm casting activity

decreased logarithmically with increasing distance from the hedgerow trees planted between crop rows, possibly due to the

greater abundance of organic matter, higher shade levels and

lower levels of disturbance beneath trees (Hauser et al. 1998). Individual tree species can differ in their effects on soil biota, due

to differences in leaf litter quality and the chemical composition of

root exudates (Campbell et al. 1994; Barrios et al. 1997; Grayston

et al. 1997; Vohland and Schroth 1999; Barros et al. 2003).

Given the potential for trees to create areas of increased soil

macrofauna activity, and the potential for soil macrofauna to

improve soil quality, research linking the presence of trees with

soil fauna activity may provide resource-poor farmers in the

tropics with an incentive to retain and plant trees within their

fields (Barrios et al. 2005 ). The presence of dispersed trees within crop fields is one of the distinctive features of the agroforestry

system used as a case study in the research reported here. The

field research took place in southern Lempira Department in

western Honduras, in a mountainous area where three-quarters of

the rural population are smallholder and subsistence farmers growing maize, sorghum and bean crops. Farmers refer to the

plots of land where they rotate these crops as mi/pa. The case

study area is notable for the fact that it represents a transition

from traditional slash-and-bum agriculture to a reportedly more

sustainable, new method of slash-and-mulch agroforestry (Hellin

et al. 1999; Ordonez Barragan 2004). Trees and shrubs that were previously killed by the fires used to convert secondary forest into

milpa plots are now retained by farmers during the process of

selectively slashing the forest to create crop plots. Trees are either

left as free-growing specimens, or as coppiced short ( < 1.5 m)

shrubs. Material from the trees is pruned and left on the ground as

mulch during the growing season. The remnant trees deliver a

range of environmental and economic benefits for farmers, such

as the supply of firewood, fruits, timber, mulch, feed for livestock,

reduction of soil erosion, greater retention of soil moisture and

provision of shade (Hellin et al. 1999; Ruben and Clercx 2003).

However, remnant trees also compete with crops for light,

nutrients and water.

Objectives

The overall aim of the research was to investigate fine-scale

temporal and spatial influences on the distribution of earthworm casting activity in agroforestry fields in an area of western

Honduras, with particular attention given to the role of trees in

influencing organic resource distribution. The first objective of the study was to determine whether structured spatial patterns exist

in the distributions of surface earthworm casts and litter. The

second objective was to relate spatial patterns in earthworm

surface activity and litter to the distribution patterns of trees

within fields, using a grid-based sampling approach. The third

objective was to determine the stability of spatial distribution

patterns of soil macrofauna over short time periods of several

weeks. Two farmers' fields were sampled in the study, one of

which was known to have a relatively high abundance and

biomass of soil macrofauna, whereas the other was known to have

a relatively sparse soil macrofauna community. A comparison of

the distribution patterns of organic resources and earthworm

surface casting (as an easily sampled indicator of earthworm

activity) between these two sites might reveal why some

agroforestry fields have a large soil macrofauna biomass, and

others do not. Such information could become a useful manage

ment tool for farmers.

Materials and methods

Study area and study sites

The study area was located in the zone surrounding the village of Candelaria, in the southern region of Lempira Department in

south-western Honduras (Fig. 1 ). Lempira is one of the least

economically developed departments of Honduras (Ruben and

Clercx 2003), due largely to its rugged topography, lack of infrastructure and isolation from the rest of the country. The

climate of the study area is classified as equatorial winter dry (Aw) according to the Koppen-Geiger classification (Kottek et al. 2006).

N. Pauli et al./ Pedobiologia 53 (2010) 127-139 129

HONDURAS

Fig. 1. Location of study area.

Annual rainfall, which falls primarily between early May and late October, averages 1200-1400 mm (Cherrett 1999). Average daily

temperatures range between 17 and 25 °C (Hellin et al. 1999).

The study area falls within the Central American dry tropical

forest zone, which has been almost completely converted to agriculture across its original extent since the initial settlement of the area over 1000 years ago Uanzen 1988; Barrance et al. 2003; Gordon et al. 2003 ). The soils of the study area are Entisols that

are generally shallow, acidic ( pH of less than 5.1 ), with low organic matter content and available phosphorus, and are mostly sandy

clay loam and clay loam in texture (Hellin et al. 1999; Ordonez Barragan 2004; Pauli 2008).

Mi/pa fields belonging to two local farmers were selected as the study sites for intensive sampling of earthworm surface casting

activity. The two sites were similar in many respects, including the number of years since conversion from secondary forest to

mi/pa, soil texture (loam to sandy loam), altitude (ranging between 500 and 600m a.s.l.) and geographic location (the two

sites are separated by ~ 500 m). However, during a prior field

survey in the wet season of 2004, the study sites recorded large

differences in the abundance and biomass of earthworms (Pauli

2008). During the 2004 field survey, ten soil macrofauna samples

were extracted along a 90 m transect placed within each site, using the standard soil macrofauna sampling method of the

Tropical Soil Biology and Fertility Institute (Anderson and Ingram 1993). In the first milpa field (referred to as the low earthworm

biomass site) a mean earthworm biomass of 6.1 ± 2.2 g m-2 was

recorded. In the second mi/pa field (referred to as the high earthworm biomass site), soil macrofauna biomass averaged

95.4±24.Sg m-2• The farmers who owned the fields also

recognised that the two sites differed in soil quality and crop yield, with the site with more earthworms generally considered to

be more fertile than the other site.

Field methods

Sampling for earthworm surface casting activity was undertaken during the wet season of 2005. One sampling grid was

placed in each of the two subject farms. Sample points within the grids were located two metres apart, using a standard 'square'

layout. Care was taken to ensure that sampling grids were located within areas that were relatively homogeneous in terms of local

soil type (as identified by the farmer) and topography. Both sampling grids were located on sloping terrain.

In the low earthworm biomass site, the sampling grid was

placed in an area identified by the farmer as below-average soil

quality, due to the fact that the soils were yellow-coloured and compacted. The sample grid measured 20 m x 20 m, and consisted

of 11 'columns' running parallel to the main slope angle by 11

'rows' running perpendicular to the main slope angle. The basic grid was sampled for litter cover and earthworm casts on two

occasions, on 7 and 29 September 2005. There were several days

of heavy rainfall associated with a tropical low pressure system prior to the second sampling event. On the second sampling date,

additional sample points at shorter intervals were placed along

the diagonals of the sampling grid to increase sampling density. In the high earthworm biomass site, the sample grid was

located in an area identified by the farmer as containing fertile,

dark-coloured soils rich in organic matter. In this site, the

sampling grid measured 16 m x 22 m, consisting of nine 'columns'

running parallel to the main slope angle and 12 'rows' running perpendicular to the main slope angle. This elongated shape was

adjusted to the topography of the site, which sloped away on either side into small gullies. The grid was sampled twice for

earthworm casts, on 8 and 22 September 2005, and once for litter cover on 8 September 2005.

Within each grid, measurements were made of earthworm

surface casting activity, litter cover and tree distribution. At each

of the sampling locations, earthworm casts with a discernible.

well-formed shape were counted and collected from within a 0.25 x 0.25 m quadrat. Earthworm casts were dried at 105 °C for 24 h and subsequently weighed. Litter cover was estimated

visually from within the same quadrat. Litter cover was initially measured only for the first sampling date at each site, as it was

thought that there would be no substantial differences in litter

cover over a short time period of 2 weeks. However, following heavy, persistent rainfall associated with a tropical low pressure system, litter cover was re-sampled on the second sampling date

at the low earthworm biomass site.

Free-growing trees and pruned (or pollarded) trees and shrubs

were also measured. At each site, the grid used for sampling


earthworm surface casting activity was extended by 2 m on all sides. The numbers of free-growing and pruned trees were then counted from within 'cells' of 4 m x 4 m, with one earthworm cast sampling point located in the centre of the cell, as shown in the example layout in Fig. 2. For the free-growing trees only, the species, trunk diameter at breast height and canopy diameter were also recorded. For data analysis, the numbers of pruned trees and free-growing trees were pooled.

Data analysis

Prior to spatial data analysis, exploratory data analysis was performed for each data set. Earthworm cast weight data were highly skewed and were square-root transformed to approximate normality. Calculation of semivariance ( y ), is adversely affected by non-normal data (Webster and Oliver 2001 ). Outliers were removed from the earthworm cast weight data set, as this can also adversely affect the calculation of semivariance (Webster and Oliver 2001 ). Due to a high proportion of zero values for earthworm casts and tree density in the low earthworm biomass site, these data were converted to binary presence/absence data sets.

For each individual data set described above, spatial autocorrelation was assessed by calculating the experimental semivariogram and Moran's l. The semivariogram depicts the degree of relatedness in the value of a particular variable amongst pairs of data points at increasing distances, or lag classes (Webster 1977; Webster and

0 Earthworm cast __ Boundary of tree sample point -- sampling grid

Additional • earthworm cast

sample points

Tree sampling grid

Example tree sampling ·cell' surrounding each earthworm cast sample point

Boundary of - earthworm r.ast

sampling grid (4 m x4 m)

. . ' . . . :.

o o ◊ ··O· 6· 6· ·O 6 6······ i • i i : • '

· -6 ··6 6 ··◊ ·◊ 6· O·· O· ··<!>···. ,. . . . . .

•! ,. ' · ·◊ ·<? ◊" ·6 o ··¢··4> ·¢ 6. . . .

·O ◊ ·◊ O· .. O· O······O ·6 ·<j>

····O!· •····••··Oj· • • ' . <p 6··0 ·O·····

i •• !•.: ·¢ <:> 6 ·O <?··◊

. . : .0 ··O · 0 · · ¢ ◊ ···0 · ·

:.• ·•.:· 6 6 6 6 Q 6 · 9 · O 0 · ·

! • ! • '

· 6 · 6-· ···6 ◊ Q ··¢ ¢, · 6 ¢ . ' . . :.. . . .

L!::==================================I

}2m

Fig. 2. Sampling grid layout for earthworm casts and trees at the low earthworm

biomass site. Earthworm casts and litter measurements were sampled at point locations (denoted by the open circles), and free-growing and pruned trees were

measured within grid cells of 4 m x 4 m diameter surrounding each sampling point

(grey square). On the second sampling date, additional earthworm cast samples

were taken along the diagonals of the grid (black circles) to increase sampling

density. The sampling grid at the high earthworm biomass site followed a similar

layout, although the dimensions of the earthworm cast sampling grid at that site

were 16m x 22 m.

Oliver 2001 ). Lag classes were defined as 2 m, which was the minimum distance between sample points. Model semivariograms were selected by eye from a choice of different models (including spherical, exponential, Gaussian and linear models) offered in the statistical package GS+ (Gamma Design Software 2001 ), and were used to assess the percentage of spatial variation in each data set that was described by the model. Spherical models can be used to describe a 'typical' pattern of spatial autocorrelation, with increasing dissimilarity between pairs of points at increasing distances and regularly sized patches of similar values (Webster and Oliver 2001 ). Exponential models tend to describe spatially autocorrelated data where the patches of similar values in space are of varying size (Webster and Oliver 2001 ). Moran's l was calculated for each lag class, and compared with the results of the semivariogram analysis.

Spatial correlation between different variables within each sampling grid was assessed using cross-semivariograms, which were calculated between the data for trees, and the data for earthwom1 casts and litter cover. Different data sets examined using crosssemivariograms must normally have the same coordinates (Webster and Oliver 2001 ). However, tree distribution was measured within 4 m x 4 m 'cells' and not at point locations, as for the other two variables (i.e., litter and earthworm casts). The sample points for litter and earthworm casts were located at the centre of the tree sampling cells, as shown in Fig. 2. Differences in variables between sampling dates were assessed visually by comparing graphs.

Results

Combining all sampling dates, earthworm casts were found at 44% of the grid points sampled in the low earthworm biomass site, and at all the grid points sampled in the high earthworm biomass site. For both sampling dates combined. the average density of earthworm casts in the low earthworm biomass site was 15.9 ± 1.7 g m-2

• The average density of earthworm casts was 9.3 ± 2.5 g m-2 on the first sampling date and 21.1 ± 2.3 g m - 2 on the next sampling date, following several days of heavy rainfall. In the high earthworm biomass site, the average earthworm cast density across both sampling dates was 267.6 ± 15.4 g m-2

, with similar weights recorded on both sampling dates.

Litter was composed of a mixture of crop stubble, tree prunings, sticks and decomposing leaves. The mean litter cover in the low earthworm biomass site on the first sampling date was 60.2 ± 2.7%, which was significantly different from the average litter cover in the high earthworm biomass site of 67.9 ± 2.9% (two-tailed t-test; P=0.042). The average litter cover recorded 3 weeks later in the low earthworm biomass site was 33.8 ± 2.7%.

In the low earthworm biomass site, trees were not distributed evenly across the sampling grid, but rather were concentrated around the extremities of the grid. The density of pruned trees within the grid was 694 trees ha - 1

, and the density of free-growing trees was 226 trees ha- 1. The six free-growing species encountered were caulote (Guazuma ulmifolia: Sterculiaceae), caoba or mahogany (Swietenia humilis: Meliaceae), carao (Cassia grandis: Caesalpiniaceae), guachipilin (Diphysa americana: Papilionaceae), laurel (Cordia alliodora: Boraginaceae) and nance (Byrsonima crassifolia: Malphigiaceae). In comparison, trees were more abundant and more evenly distributed across the sample grid in the high earthworm biomass site. The total density of pruned trees within the grid in the high earthworm biomass site was 2788 trees ha- 1, and the density of free-growing trees was 134 trees ha- 1

• The five free-growing species encountered within the grid at this site were cangrejillo ( Cupania sp.: Sapindaceae), caoba, chichipate (Lonchocarpus sp.: Papilionaceae), guachipilin and silimera (Poeppigia procera: Caesalpiniceae).


Spatial autocorrelation in measured variables

For the low earthworm biomass site, the litter cover values changed substantially in the 3 weeks between the two sampling events (Fig. 3). During the first sampling event, there was clear spatial autocorrelation for litter cover, with positive values of Moran's I at distances of less than 5 m, and spatial autocorrelation present at distances less than 10 m ( i.e., the range). At distances of greater than 10 m, the semivariance values decreased, introducing noise into the spatial pattern. A spherical model fitted to the semivariogram data explained around 42% of the total spatial variation in litter cover. A spherical model would indicate that patches of litter agglomerations are roughly similar in size throughout the sampling grid.

The semivariogram and correlogram from the second litter sampling event at this site also showed clear spatial autocorrelation in litter cover values, with an increased patch size compared to the first sampling event (Fig. 3). Patches of litter with similar cover values tended to be around 8 m in diameter. The spherical model fitted to the semivariogram data explained around 58% of

First sampling event (7" September 2005)

Litter distribution

·········1·

•.•..•••• IIH•·•••·· ······•• • •••••••••• •• • • • • •••........... •····•····••••••••••••••••••••••••••••••••••

6

., -�

4

0

0

1,0

o.s

-�� o.o

::E

-0.5

,t.0

Semlvariogram. first event

•

• •

10 15

Separation dislance (m)

Corre/ogram • first event

Separation distance (m)

the total variation in litter coverage, with spatial autocorrelation in litter values predicted up to a range of 17.9 m.

At the high earthworm biomass site, litter cover was fairly even across the sampling grid, with increasing dissimilarity between litter cover values at pairs of points with increasing distance (Fig. 4). The semivariogram was unbounded, which means that an upper limit to the distance at which litter values are no longer spatially autocorrelated was not reached within the sampled area.

The distributions of earthworm cast weights on both sampling dates for the low earthworm biomass site are shown in Fig. 5. Earthworm cast weights were sampled 3 weeks apart at this site. Earthworm casts were widely dispersed on the first sampling date, and the semivariogram and correlogram (Fig. 5) indicate that there was little spatial autocorrelation in the data at the scale measured. On the second sampling date, earthworm cast weights were much greater, and the number of grid points with earthworm casts also increased. Although the distribution diagram appears to show areas of patches and gaps· of earthworm surface casting activity, the semivariogram and

Second sampling event (29., September 2005)

Litter distribution

··•· ...... ·•

··r.· .• . . . . " • • • • ••••.• . . .. "· ·•

• •

. . .•. . ., ........

-=�� . ..... ... . .•·••·1·· ...... ..

10 Semlvariogram • second event

8

-� 8

-�4

0

s 10 15


t.O Correlor,ram • sea,nd event

0.6

• •

i •

• 00

• ::i

10 115

-0.5

-1.0


Fig. 3. Spatial distribution of litter cover at the low earthworm biomass site, together with corresponding semivariograms and correlograms. The upper diagrams show the

spatial distribution of litter cover on the first and second sampling dates, 3 weeks apart. The relative size of the black circles indicates the percentage litter cover; the range

of values was 5-100% for the first event, and 0-90% for the second event. The middle and lower diagrams show the semivariograms and correlograms (Moran's/) for litter

cover at each of the sampling dates. The solid line on the semivariogram for the first sampling date corresponds to a fitted spherical model with the following parameters:

nugget=3.011; sill=5.520; range=10.30m; proportion of variation explained: 0.415. The solid line on the semivariogram for the next date corresponds to a fitted spherical

model with the following parameters: nugget= 3.340; sill= 7.935; range= 17.9 m; proportion of variation explained: 0.579.

132 N. Pauli et al./ Pedobio/ogia 53 (2010) 127-139

Litter distribution

••••••••• 10

Semivariogram

••••••••• Q)

8 •

•• •

• C: • -� 6 •

••••••• • • -� 4

••••••••• 2

••••••••• 0

••••••••• 0 5 10 15


••••••••• 1.0 Correlogram

• • ••••••••••••••••

0.5

_v, •

••••••• C: •

• • � 0.0

5 10 15

• ··•··• ::ii

• • -0.5

·····••· • -1.0


Fig. 4. Spatial distribution of litter cover at the high earthworm biomass site. and corresponding semivariogram and correlogram. The litter distribution diagram on the left

shows the coverage of litter within the site on the first sampling date (8 September 2005 ). Litter was sampled only once at this site. The relative size of the black circles

indicates the percentage litter cover; values ranged between 5% and 100% cover. The semivariogram appears to be unbounded within the studied area.

correlogram did not show any distinct spatial structure in the data.

The distribution of earthworm cast weights at the surface for the high earthworm biomass site is depicted in Fig. 6. Earthworm cast weights were sampled 2 weeks apart. Although the average earthworm cast weight did not change substantially between the two sampling dates, the distribution diagrams indicate that the relative distribution of earthworm cast weights did change between sampling dates, with a greater concentration in the upslope portion of the sampling grid on the second sampling date. This overall gradient in earthworm casting activity on the second sampling date was reflected in the correlogram, which indicated positive spatial autocorrelation at distances of less than 8 m. The semivariogram showed a steadily increasing level of dissimilarity between pairs of points, with a projected range (according to a spherical model fitted to the points) of around 40 m. It is likely that the area sampled was not sufficiently large to detect the upper boundary of spatial autocorrelation in earthworm cast distribution.

Spatial cross-correlation in measured variables

Fig. 7 shows the spatial distribution patterns of litter cover in relation to trees for the two sampling events in the site with low earthworm biomass. For the first sampling event, litter cover tended to be greatest around the periphery of the sampling grid, which is also where trees were concentrated. However, the upper, central part of the sampling grid had high litter cover, but no tree cover. The cross-semivariogram indicated a high level of spatial cross-correlation between the two variables, with spatial interdependence at distances of up to 13 m. A spherical model fitted to the data explained around 87% of the total variation in cross-semivariance values. In the second sampling event 3 weeks later, litter cover was concentrated in the lower half of the sampling grid (Fig. 7). There was still a strong spatial crosscorrelation between litter cover values and tree presence, but the

range increased substantially to over 20 m compared with the first sampling event.

Visual comparison of the distribution of litter cover and tree cover in the site with high earthworm biomass did not indicate any obvious spatial relationship (Fig. 8). However, the cross-semivariogram indicated a strong degree of spatial interdependence, particularly at distances of less than 5 m. An exponential model variogram was the best fit for the data points, which may indicate that the spatial relationship between the two variables was irregular; that is, that the size of agglomerations of trees and litter within the sampling grid was variable. There appears to be a short-range structure in the data at distances of less than 5 m, and a longer-range structure at distances of greater than 5 m. The sharp incline on the cross-semivariogram and the lack of points at distances of less than 2 m means that it is difficult to judge how well the exponential model reflects the actual relationship between trees and litter at this site.

Fig. 9 illustrates the spatial relationship between trees and earthworm casts in the site with low earthworm biomass. For the first sampling date, earthworm casts were sparsely distributed throughout the sampling grid, and were most commonly encountered around the periphery of the quadrat. Trees were also most abundant around the periphery, and the similarities in the distributions of the two variables are shown in the crosssemivariogram. The spherical model variogram explained over 90% of the total variation in the semivariance values. The spatial cross-correlation between the two variables was evident up to distances of 8.5 m. For the next sampling date, a similar overall relationship between trees and earthworm casts was evident, with the main difference being an increased density of surface casts. An exponential model best described the crosssemivariance data.

The distribution patterns of earthworm casts in relation to trees in the high earthworm biomass site are depicted in Fig. 10. For the first sampling date, an exponential model variogram provided the best fit to the data points, similar to the results for litter cover and tree distribution from the same sampling event. There appeared to be two spatial structures in the data - one at


Earthworm cast distribution• first event

0 0 0 0 0 • 0 0 0 0

0 0 0 0 0 0 • 0 0- 0

• -0 -0 -0 0 0 0 0 0 • •

• • 0 0 0 0 0 • 0 •

• • 0 • 0 Q • • 0 • 0

0 0 0 0 0 • • 0 0 . 0

0 (> -0 -0 0 0 0 • • •

• • 0 0 0 0 0 0 0 0 •

-0 0 0 0 0 0 0 0 0 -0 0

-0 0 • 0 • • 0 -0 • • •

• -0 • • 0 • • • • 0 0

O.◄ Semivariogram • first event

0.3

g.. • • • • • •

·� 0.2 •

0,1

0.0

0 5 10 15 Separation distance (m)

1.0 Correlogram • first event

0.5

., •

-� 0.0 ■s ::. 10

-0.5

·1.0 Separation distance (m)

Earthworm cast distribution - second event

..... 0 0 ••• •

,

•K• .•....• • • o o • ·o• e.

••••• 0 ♦ • o0

0 ••

···•·•••

0o

•o o ••

•• 0 0 0 :.;• 0 •••

•• 0 •

•••• • 0 • • •

• · 0 ·0• •• 0 •• o e e

• 0 o0

0 •• 0 0 ••••

• O

O O O O - 0 • 0 •

0

• ♦

•

o•

e o e o o • • 0 • 0

e

.I

.;,

·"'� ::.

o.s Semlvarlogram • second event

o.•

0.3

0.2

0.1

0.0 0

1.0

05

0.0

-0.5

.,.o

••

• • • • • •

5 10 15 Separation d'<stance (m)

Correlogram • second event

■


Fig. 5. Spatial distribution of earthworm surface casting activity at the low earthworm biomass site, and corresponding semivariograms and correlograms. For the

distribution diagrams at top, the circles represent the relative weight of earthworm casts collected from each sampling point. The range in cast weight was 0-16.6 g for the first sampling event, and 0-10.6 g for the second sampling event. The small open circles denote grid sample points where no earthworm casts were recorded. Grid size was

20 m x 20 m, with samples every 2 m. The semivariograms and correlograms for each sampling event are shown beneath the litter distribution diagrams. Model semivariograms were not fitted, as there was no obvious spatial autocorrelation present.

short distances of less than 4 m, and another longer-range structure. Similarly to litter cover, the sharp incline on the cross-semivariogram and the lack of points at distances of less than 2 m means that it is difficult to judge how well the

exponential model reflects the actual relationship between trees and earthworm cast distribution. On the second sampling date (Fig. 10), there was clear spatial cross-correlation between earthworm cast distribution and tree distribution. This was evident on the cross-semivariogram, with increasing dissimilarity between the values of earthworm casts and trees up to around 8.5 m. The spherical model variogram explained

around 71.2% of the total variation in the cross-semivariance values.

Temporal variability in distribution of litter and earthworm casts

Litter was sampled twice at the low earthworm biomass site. The distribution graphs clearly show the change in litter

distribution between the two sampling events (Fig. 3). Before the unusually heavy rainfall event, litter cover was fairly evenly

distributed across the sampling grid, but after the rainfall event, litter cover was higher in the lower half of the sampling grid than in the upper half.

As previously noted, there was a marked difference in earthworm cast weight and distribution in the low earthworm biomass

site between the two sampling dates. The mean cast weight was

three times greater for the second sampling event than for the first sampling event, and the number of point locations at which earthworm casts were found doubled between the two sampling events. At the high earthworm biomass site, mean earthworm cast weight was similar for both sampling events, although it should

be noted that the spatial relationship between trees and earthworm casts was stronger on the second sampling date than on the first (Fig. 10).

Discussion

Spatial patterning of litter and earthworm casts

With some notable exceptions, the variables investigated did not exhibit strong spatial autocorrelation at the scale investigated.


Earthworm cast distribution • first event

• • • • • • • • • •• • • • ••• •••• •• . ·- •

• • • • • • • • • • • • • • •• • •

• • • • • ••· • • • • • ••••• • • •••••••• • • • • • • • • • • • • •· • • • • •• • • ••· •

• •• • • ••• •3 Semivariogram • fitst event

" u 2 • ■ ■ C

• • ■

.:g •

"' -�

{/)

0

0 5 10 15


1.0 Co"elogram - first event

0.5

-�s 0.0

5 10 15 :::;;

-0.5

-1.0


Earthworm cast distribution • second event

•••••••••···•· ·•1·••• • • • • • •

-�r!0

:::;;

• • • • • • • • •

• •• •••• • • • • • • • • • • •

• • • • • • • • • • • • • • • • • •

•

•

<D g 2

-�

-�

� 1 {/)

0

1.0

0.5

0.0

-0.5

-10

• • • • • • ••

• ••• • •

• • • • • • • •

• • • ·••· •

Semfvar/ogram • second event

.-----

0 5 10 15


Co"elogram - second event

■ •■

5 5


Fig. 6. Spatial distribution of earthworm surface casting activity at the high earthworm biomass site, and corresponding semivariograms and correlograms . Earthworm cast

weights were sampled twice, 2 weeks apart . For the distribution diagrams at top, black circles represent the relative weight of earthworm casts at each sample site; sizes

are comparable between graphs. The range in cast weight was 1.3-76.0g in the first week, and 0.6-90.6g in the second week. Grid size was 16 m x 22 m, with grid points

spaced 2 m apart. In the first sampling event, no obvious spatial trend was evident, as shown by the semivariogram and correlogram. In the second event, a spherical model

was fitted to the semi variance data with the following parameters: nugget= 1.655; sill=3.543; range= 39.53 m (not reached within sampling grid); proportion of variation

explained: 0.534.

The strongest spatial autocorrelation was found for litter distribu

tion in the low earthworm biomass site on both sampling events

(i.e. before and after the extreme rainfall event). The main

difference in the spatial distribution of litter between these two

dates was that the distance over which autocorrelation occurred

increased from around 10 m before the rainfall event to almost

18 m after the event. This suggests the possibility that a different

set of environmental variables was driving the distribution of

litter on the two sampling dates.

There were few clear patterns of spatial autocorrelation in the

earthworm cast data collected. Spatial autocorrelation may have

been more notable if a larger area had been sampled, as indicated

by the unbounded semivariograms for earthworm cast distribu

tion on the second sampling date in the site with high earthworm

biomass. Over scales of 20-80 m, spatial aggregation of earth

worm casting activity may be more apparent than at distances of

less than 20 m. Other studies of the spatial distribution of

earthworms have typically placed sample points 5 or 10 m apart.

to cover a maximum distance of 45-90 m. 'Patches' and 'gaps' of

earthworm density have been found at these scales (Rossi 2003a,

2003b, 2003c). As ea1thworms are relatively large, mobile soil

invertebrates, their distribution at scales of 20-80 m may be

controlled by demographic processes rather than by resource

distribution.

Spatial relationships between earthworm surface casting, litter and

trees

When considered alone, none of the earthworm cast data

collected at either site on either sampling date showed strong

spatial patterning at scales of less than 20 m. However, with the

inclusion of trees as a co-variable in the analysis, a strong spatial

structure emerged in the distribution of earthworm casts for all


First sampling event

Second sampling event

• • • • •

• Litter cover

E> Pruned trees

OFree growing trees

40

30

Cross-semlvariogram

0 5 10 15


Cross-semivariogram

0 5 10 15


Fig. 7. Comparison of litter cover and tree distribution, low earthworm biomass site. The large diagrams show the distribution of pruned and free-growing trees superimposed on the distribution of litter cover (range of values for litter cover for first event: 5-100%; for second event: 0-90%). The size of the filled grey circles indicates

the number of pruned trees found within each sampling cell (range of values: 1-3 pruned trees). The size of the open grey circles represents the size of the tree canopy. The

smaller diagram at bottom right for each sampling date shows the cross-semivariogram for the spatial relationship between tree distribution and litter distribution. For the

first sampling event, a spherical model semivariogram was fitted with the following parameters: nugget=4.15; sill=32.09; range= 13.44 m; proportion of variation explained=0.871. For the second sampling event, a spherical model semivariogram was fitted with the following parameters: nugget=6.38; sill=30.55; range=22.18 m (not reached within sampled area); proportion of variation explained: 0.791.

four comparisons. This may indicate that tree distribution is a

strong driver of the spatial distribution of earthworm surface

casting at the scale investigated. For two of the four comparisons,

a spherical model with a range of around 8.5 m was the best fit.

This may indicate that trees have a broader 'zone of influence' than just the canopy dripline (average canopy width was around

4 m in both sample grids), which could be explained by the lateral

extent of root networks around trees. Alternatively, there may also

be a degree of spatial aggregation in tree distribution, resulting in 'clumps' of trees and earthworm casting activity. Tree distribution

and density would need to be mapped over a larger area (perhaps

100 m x 100 m) to assess whether such patterns of spatial

aggregation exist within fields. For the other two comparisons,

an exponential model provided the best fit to the cross

semivariance data, which may indicate that the width of spatial

aggregations of trees and earthworm casts was not constant

throughout the sampling grid. For one of these comparisons there

appeared to be a 'short-range' spatial structure at distances of less

than 4 m, where cross-semivariance increased steeply over this

distance. This 'short-range' structure may be indicative of the

zone of influence of individual trees, whereas the 'long-range'

spatial structure at greater distances may be related to spatial

aggregation in litter or tree distribution. Further observations

would be required to provide support for this hypothesis,

including observations at shorter sampling distances.

In both sites, there was a strong positive association between

tree distribution and litter cover. In the low earthworm biomass

site, the distance over which this association occurred increased

from around 13 m before the rainfall event, to over 20 m after

heavy rainfall. The exponential form of the variogram in the high earthworm biomass site indicated that the cross-semivariance

increased sharply up to 5 m, and then gradually levelled off. As for


• Litter cover

0 Pruned trees

0 Free growing trees

50

40

30

20

10

Cross-semivariogram

0 +----...-----.----,

0 5 10 15


Fig. 8. Comparison of litter cover with tree distribution, high earthworm biomass site. The large diagram shows the distribution of pruned and free-growing trees superimposed on the distribution of litter cover (range of values for litter cover: 5-100%). The size of the filled grey circles indicates the number of pruned trees found within each sampling cell (range of values: 1-4 pruned trees). The size of the open grey circles represents the size of the tree canopy. The smaller diagram at bottom right shows the cross-semivariogram for the spatial relationship between tree distribution and litter distribution. An exponential model variogram provided the best fit to the data points, with the following parameters: nugget=3.0; sill=39.68; range=7.15 m.

earthworm cast distribution in this site, this could indicate that there are different drivers of the aggregation of litter cover and tree density at short distances than over longer distances. The association of litter cover and tree distribution was expected,

as farmers tend to let mulch fall on the ground in the immediate vicinity of the pruned tree.

Although a strong positive relationship was observed between earthworm casts and tree distribution, this study did not investigate any of the mechanisms by which these relationships may have been generated. Earthworm casting activity is concen

trated in the areas where earthworms are feeding (Lavelle et al. 2001; Jimenez and Decaens 2004; Mariani et al. 2007). A preference for feeding beneath trees may be due to

the increased organic matter often found in these areas, or to greater soil moisture retention (Coleman et al. 1991; Hauser et al. 1998; D0ckersmith et al. 1999). Of course, there exists the possibility that the distributions of trees and soil macrofauna were controlled by an additional, unsampled variable (Saetre and Baath 2000).

Temporal change in spatial distribution patterns

The passage of an unusual weather event led to the investigation of temporal stability of litter distribution in one of the two study sites. Litter distribution in this site was substantially modified following the heavy rainfall. The distance over which litter and tree distribution were spatially correlated increased from around 13 m before the rainfall event, to over 20 m after heavy rainfall. Likewise, spatial autocorrelation patterns of litter distribution increased from 10 m before the rain, to 18 m after the event. One possible explanation is that litter distribution was closely related to tree location under 'normal' weather conditions, but after heavy rainfall, surface run-off caused litter to be moved down slope, so that position on slope was a more important driver

of litter distribution than tree location. Several local farmers stated that the wet season rains of 2005, which included rainfall associated with Hurricane Stan in late September and early October, were the heaviest in many years. Within the study area,

Hurricane Stan was generally considered to have been worse than the highly destructive Hurricane Mitch of 1998, in terms of the amount of rain that fell and the adverse effects on crops.

Earthworm cast distribution and mass changed substantially in one of the sites, where casts were measured before and after the heavy rainfall event, but not in the other site, where casts were only

measured during typical wet season conditions. In both sites, the distance over which trees and earthworm cast weight were spatially correlated was greater during the second sampling event than during the first sampling event. In the low earthworm biomass site, this increase occurred after heavy rainfall. It is possible that this increased distance is related to change in litter distribution patterns after rainfall, changes in spatial patterns of belowground organic resources, or to the effects of topography or another variable that may be an important driver of earthworm distribution at broader scales. Because data on earthworm cast distribution were not

available from both sites for all three sampling dates, it is not possible to determine whether the patterns displayed in each site are representative of more general changes in earthworm surface casting activity after heavy rains. The fact that the most substantial temporal changes were the only two measured before and after the heavy rainfall suggests that this weather event did play a role in modifying soil biological properties that otherwise might not have been so variable over short time periods.

Differences between the two sites

The distance over which earthworm casts and litter were spatially correlated was greater in the site with lower earthworm biomass, which also had lower tree density and soil quality than

..

► N. Pauli et al./ Pedobiologia 53 (2010) 127-139 137


0 • 0 0

0 • 0

0 0 0 0

0 0 0 0 •

0 0 0 • 0

e 0 • • 0 0


-..1-·--• 0 0 •

-◊

··

• 0 0

◊. 0 e • o0

·u

····o • o•o

• • 0 @.

0 •

• Earthworm cast weight

0 Pruied trees

0 Sample with no earthworm casts 0 Free growing trees

0.15 Cross-semivariogram

0 5 10 15


e•. 0 0 0 :◊�0 • 0 •

• 0 -,

◊·· 0 .•.

• • • 0

, •o 0 ◊ .

.

£.. • 0 0 ·◊

1.2

G)

g 0.9 -�

Cross-semivariogram

0 0-0 • 0.

0 • . . . 0

() 0

-� 0.6

� 0.3

0 5 10 15


Fig. 9. Comparison of earthworm casts and tree distribution, low earthworm biomass site. The large diagrams show the distribution of pruned and free-growing trees superimposed on the distribution of earthworm surface casts for each sampling date (range of values for earthworm casts on first sampling date 0-16.62 g; range of values for earthworm casts on second sampling date: 0-10.62 g). The size of the filled grey circles indicates the number of pruned trees found within each sampling cell (range: 1-3 pruned trees). The size of the open grey circles represents the size of the tree canopy. The smaller diagrams at bottom right show the cross-semivariogram for the spatial relationship between tree distribution and earthworm cast distribution on each sampling date. For the first date. a spherical model variogram provided the best fit to the data points. with the following parameters: nugget=0.01; sill=0.12; range=S.5 m: proportion of variation explained=0.917. For the second date. an exponential model variogram provided the best fit to the data points, with the following parameters: nugget=0.03; sill= 1.22; range= 10.36 m.

the site with higher earthworm biomass. The 'zone of influence' of

each tree appears to be larger in the lower soil quality site. One

possible explanation for these patterns is that the greater

patchiness of trees and litter in the low earthworm biomass site

has led to the concentration of earthworm activity in small areas

where relatively scarce organic resources are most abundant.

There may also be important differences in tree root networks and

belowground production between the two sites. For instance,

Casper et al. (2003) found that the lateral extent of tree root

networks was greater in coarse-textured soils than in fine

textured soils.

Conclusion

At the scale investigated, the spatial distribution of litter cover

showed strong trends of aggregation, whereas the spatial patterns

of earthworm cast weight did not exhibit significant trends of

spatial autocorrelation. The spatial distribution patterns of both

litter cover and earthworm casts variables were well explained by

the presence of trees. Both earthworm surface casting activity and

litter cover were positively associated with tree distribution.

These results suggest that farmers may be able to manipulate

populations of earthworms indirectly by managing the density

and spatial distribution of trees within their crop fields. However,

there will be a trade-off between the potential positive effects of trees, litter and earthworms on soil quality, and the potential

negative effects on crop yield. Large trees have a multitude of

economic and environmental benefits for farmers, but they also

compete with crops for light, water and nutrients. Spatial

manipulation of tree distribution could allow farmers to create

patches of organic resources within fields that act as 'source areas'

for earthworms and other beneficial soil fauna species, while at

the same time minimising the effect of trees on crop yield, and




• Earthwonn cast weight

O Pruned trees

OFree growing trees

2.5

2.0

1.5

1.0

0.5

Cross-semivariogram

0.0 ----,---�--�

3

0 5 10 15 Separation distance (m)

Cross-semivariogram

_.....,.. __ .........

0 +---�---�--�

0 5 10 15


Fig. 10. Comparison of earthworm casts and tree distribution, high earthworm biomass site. The large diagrams show the distribution of pruned and free-growing trees superimposed on the distribution of earthworm surface casts (range of values for earthworm casts for first sampling date: 1.3-76.0 g; for second sampling date: 0.6-90.6 g).

The size of the filled grey circles indicates the number of pruned trees found within each sampling cell (range of values: 1-4 pruned trees). The size of the open grey circles represents the size of the tree canopy. The smaller diagrams at bottom right show the cross-semivariogram for the spatial relationship between tree distribution and

earthworm cast weight for each sampling date. For the first date, an exponential model variogram provided the best fit to the data points, with the following parameters:

nugget=0.01: sill =2.15; range= 11.35 m. A spherical model variogram was fitted to the data points from the second sampling date, with the following parameters:

nugget=0.7, sill=2.427, range=8.5 m, proportion of variation explained=0.712.

balancing the need for an even distribution of mulch across crop

fields. Retaining trees within fields of relatively low soil quality

may be particularly beneficial in promoting soil fauna activity,

given the greater areal extent of the spatial correlation between

tree distribution and earthworm casting activity in the site with

low earthworm biomass.


Acknowledgements

The authors wish to thank Don Miguel Angel Cruz and Don

Lindolfo Arias for allowing their fields to be sampled for this study. The authors are grateful to: Edwin Garcia, Daniel Vasquez

and Qdvin Ayala for their contributions as field assistants in Candelaria; Marco Tulia Trejo of CIAT Honduras for logistical

support; Dr. Karen Holmes for constructive comments on the research; staff at ClAT Cali, ClAT Honduras, FAQ Candelaria, FAQ

Tegucigalpa and the School of Earth and Environment at the University of Western Australia; and two anonymous reviewers for their constructive criticism of the manuscript. Financial

support was provided by Centro Internacional de Agricultura

Tropical (CIAT), the School of Earth and Environment at the University of Western Australia, and an Australian Postgraduate

Award to the first author.

References

Anderson, J .. 1988. Spatiotemporal effects of invertebrates on soil processes. Biol. Fertil. Soils 6, 216-227.

Anderson,j.. lngram,j.. 1993. In: Tropical Soil Biology and Fertility: A Handbook of Methods. CAB International. Wallingford. UK.

Barrance, A., Flores. L. Padilla, E., Gordon, J., Schreckenberg, K., 2003. Trees and farming in the dry zone of Honduras I: Campesino tree husbandry practices. Agroforest. Syst. 59, 97 -106.

Barrios, E., 2007. Soil biota, ecosystem services and land productivity. Ecol. Econ. 64, 269-285.

Barrios, E .. Cobo, J.G., Rao, J.M., Thomas, R.J., Amezquita, E., Jimenez, J.J., Rondon, M., 2005. Fallow management for soil fertility recovery in tropical Andean agroecosystems in Colombia. Agric. Ecosyst. Environ. 110, 29-42.

Barrios, E .. Kwesiga, F., Buresh, R.j., Sprent, J.1., 1997. Light fraction soil organic matter and available nitrogen following trees and maize. Soil Sci. Soc. Am. J. 61, 826-831.

Barros, E., Neves, A., Blanchart, E., Fernandes, E.C., Wandelli, E., Lavelle, P., 2003. Development of the soil macrofauna community under silvopastoral and agrosilvicultural systems in Amazonia. Pedobiologia 47, 273-280.

Beare, M., Coleman, D., Crossley, D.J., Hendrix, P., Odum, E., 1995. A hierarchical approach to evaluating the significance of soil biodiversity to •biogeochemical cycling. Plant Soil 170, 5-22.

Bhadauria, T., Ramakrishnan, P., 1989. Earthworm population dynamics and contribution to nutrient cycling during cropping and fallow phases of shifting agriculture Uhum) in north-east India. J. Appl. Ecol. 26, 505-520.

Campbell, B., Frost, P., King, J., Mawanza, M .. Mhlanga, L., 1994. The influence of trees on soil fertility on two contrasting semi-arid soil types at Matopos, Zimbabwe. Agroforest. Syst. 28, 159-172.

Casp_er. B.B .. Schenk, H.J., Jackson, R.B., 2003. Defining a plant's belowground zone of influence. Ecology 84, 2313-2321.

Cherrett, I., 1999. Soil conservation by stealth. International Agricultural Development 19, 13-15.

Coleman, D., Edwards, A., Belsky, A., Mwonga, S., 1991. The distribution and abundance of soil nematodes in East African savannas. Biol. Fertil. Soils 12, 67-72.

Decaens, T., Rossi, J.-P., 2001. Spatio-temporal structure of an earthworm community and soil heterogeneity in a tropical pasture (Carimagua, Colombia). Ecography 24, 671-682.

D0ckersmith, J.C., Giardina, C.P .. Sanford, R.L., 1999. Persistence of tree related patterns in soil nutrients following slash-and-burn disturbance in the tropics. Plant Soil 209, 137-156.

Ettema, C.H., Wardle, D.A., 2002. Spatial soil ecology. Trends Ecol. Evol. 17, 177-183. Gamma Design Software, 2001. GS+ Geostatistics for the Environmental Sciences.

Gamma Design Software, Plainwell. Gordon, J., Hawthorne, W., Sandoval. G., Barrance, A., 2003. Trees and farming in

the dry zone of southern Honduras II: the potential for tree diversity conservation. Agroforest. Syst. 59, 107-117.

Grayston, S.J., Vaughan, D., Jones, D .. 1997. Rhizosphere carbon flow in trees, in comparison with annual plants: the importance of root exudation and its impact on microbial activity and nutrient activity. Appl. Soil Ecol. 5, 29-56.

Hauser, S., Asawalam, D., Vanlauwe. B., 1998. Spatial and temporal gradients of earthworm casting activity in alley cropping systems. Agroforest. Syst. 41, 127-137.

Hellin, J., Welchez, L.A., Cherrett. I., 1999. The Quezungual System: an indigenous agroforestry system from Western Honduras. Agroforest. Syst. 46, 229-237.

Janzen, D., 1988. Tropical dry forests: the most endangered major tropical ecosystem. In: Wilson, E. (Ed.), Biodiversity. National Academy Press, Washington, DC, pp. 130-137.

Jimenez, J., Decaens, T., 2004. The impact of soil organisms on soil functioning under neotropical pastures: a case study of a tropical anecic earthworm species. Agric. Ecosyst. Environ. 103, 329-342.

Jimenez,J.J., Rossi,J.-P., Lavelle, P .. 2001. Spatial distribution of earthworms in acidsoil savannas of the Eastern Plains of Colombia. In: Jimenez, J.J., Thomas, R.J. (Eds.), Nature's Plow: Soil Macroinvertebrate Communities in the Neotropical Savannas of Colombia. Centro lnternactional de Agricultura Tropical, Cali, pp. 104-120.

Kottek, M .. Grieser,J., Beck, C., Rudolf, B., Rubel, F., 2006. World Map of the KoppenGeiger climate classification updated. Meteorol. Z. 15, 259-263.

Lavelle, P., Bignell, D., Lepage, M., Walters, V .. Roger, P., Ineson, P., Heal, 0., Dhillion, S., 1997. Soil function in a changing world: the role of invertebrate ecosystem engineers. Eur. J. Soil Biol. 33, 159-193.

Lavelle, P., Barros, E., Blancha rt, E., Brown, G., Desjardins, T .. Mariani, L.. Rossi, J.P .. 2001. SOM management in the tropics: why feeding the soil macrofauna?. Nutr. Cycl. Agroecosys. 61, 53-61.

Lee, K., Foster, R., 1991. Soil fauna and soil structure. Aust. J. Soil Res. 290, 745-775.

Legendre, P., Dale, M.R., Fortin, M.J .. Gurevitch, J., Hohn, M .. Myers, D .. 2002. The consequences of spatial structure for the design and analysis of ecological field surveys. Ecography 25, 601-615.

Mariani, L.,Jimenez,J.J .. Asakawa, N., Thomas, R.J .. Decaens, T .. 2007. What happens to earthworm casts in the soil? A field study of carbon and nitrogen dynamics in Neotropical savannahs. Soil Biol. Biochem. 39, 757-767.

Mordelet, P .. Abbadie, L, Menaut, J., 1993. Effects of tree clumps on soil characteristics in a humid savanna of West Africa (Cote d'Ivoire). Plant Soil 153, 103-111.

Netuzhilin, I., Chacon, P., Cerda, H .. Lopez-Hernandez, D., Torres, F .. Paoletti, M.G .. 1999. Assessing agricultural impact using ant morphospecies as bioindicators in the Amazonian savanna-forest ecotone (Puerto Ayacucho, Amazon State), Venezuela. In: Reddy, M.V. (Ed.), Management ofTropical Agroecosystems and the Beneficial Soil Biota. Science Publishers Inc., Enfield, pp. 288-350.

Ordonez Barragan, J., 2004. Main factors influencing maize production in the Quesungual agroforestry system in Southern Honduras: an explorative study. M.Sc. Thesis, Wageningen University, Waginengen.

Palm, C., 1995. Contribution of agroforestry trees to nutrient requirements of intercropped plants. Agroforest. Syst. 30, 105-124.

Pandey, A., Palni, L.M.S., 2007. The rhizosphere effect in trees of the Indian central Himalaya with special reference to altitude. Appl. Ecol. Environ. Res. 5, 93-102.

Pauli, N., 2008. Environmental influences on the spatial and temporal distribution of soil macrofauna in a smallholder agroforestry system of western Honduras. Ph.D. Thesis, University of Western Australia, Perth.

Phillips, R.P., Fahey, T.J., 2006. Tree species and mycorrhizal associations influence the magnitude of rhizosphere effects. Ecology 87, 1302-1313.

Rhoades, C., 1997. Single-tree influences on soil properties in agroforestry: lessons from natural forest and savanna ecosystems. Agroforest. Syst. 35, 71-94.

Rossi, J.P., 2003a. Short-range structures in earthworm spatial distribution. Pedobiologia 47, 582-587.

Rossi, J.P., 2003b. Clusters in earthworm spatial distribution. Pedobiologia 47, 490-496.

Rossi, J.P., 2003c. The spatiotemporal pattern of a tropical earthworm species assemblage and its relationship with soil structure. Pedobiologia 47, 497-503.

Rossi, J.-P., Blanchart, E .. 2005. Seasonal and land-use induced variations of soil macrofauna composition in the Western Ghats, southern India. Soil Biol. Biochem. 37, 1093-1104.

Rossi, J.P., Lavelle, P., 1998. Earthworm aggregation in the Savannas of Lam to (Cote D'Ivoire). Appl. Soil Ecol. 7, 195-199.

Ruben, R .. Clercx, L., 2003. Rural finance, poverty alleviation and sustainable land use: the role of credit for the adoption of agroforestry systems in Occidental Honduras. J. Microfinance 5, 1-24.

Saetre, P., Baath, E .. 2000. Spatial variation and patterns of soil microbial community structure in a mixed spruce-birch stand. Soil Biol. Biochem. 32, 909-917.

Schooley, R.L., Wiens, J.A., 2003. Spatial patterns, density dependence, and demography in the harvester ant, Pogonomynnex rugosus, in semi-arid grasslands. J. Arid Environ. 53, 183-196.

Sileshi, G., Mafongoya, P., 2006. Variation in macrofauna communities under contrasting land use systems in eastern Zambia. Appl. Soil Ecol. 33, 49-60.

Six, J., Bossuyt, H., Degryze, S., Denef, K., 2004. A history of research on the link between (micro)aggregates, soil biota and soil organic matter dynamics. Soil Till. Res. 79, 7-31.

Vohland, K .. Schroth, G., 1999. Distribution patterns of the litter macrofauna in agroforestry and monoculture plantations in central Amazonia as affected by plant species and management. Appl. Soil Ecol. 13, 57-68.

Webster. R .. 1977. Quantitative and Numerical Methods in Soil Classification and Survey. Oxford University Press, Oxford.

Webster, R., Oliver, M., 2001. Geostatistics for Environmental Scientists. John Wiley & Sons, New York.

Zinke, P., 1962. The pattern of influence of individual forest trees on soil properties. Ecology 43, 130-133.

fine-scale spatial and temporal variation in earthworm...

Documents